"Talk to your customers" is the most repeated piece of startup advice and one of the least actionable. I have told founders this myself, in exactly those words, and then watched them come back from ten customer conversations having learned almost nothing useful.
The problem is not that founders do not talk to customers. Most founders with engaged investors talk to plenty of customers. The problem is that the conversations are structured in a way that systematically generates confirmation rather than information. Founders ask questions that lead to validation. Customers, being polite, provide validation. The founder leaves feeling encouraged. The startup continues in the wrong direction.
I ran customer development at three companies and watched it go wrong at a dozen more. Here is what I learned about why the standard advice fails, and the exact five-question structure that replaces it.
Why the Standard Advice Fails
"Talk to customers" fails for four structural reasons.
You ask about the future. "Would you use a tool that does X?" "How much would you pay for Y?" Questions about hypothetical future behavior are answered by the customer's desire to be helpful, not by their actual preferences or history. Everyone would use a tool that does X if X sounds useful enough. That tells you nothing.
You describe your solution too early. The moment you mention what your product does, the conversation becomes an evaluation of your idea rather than an exploration of the customer's reality. Customers who like you will be generous. You will mistake generosity for signal.
You talk to the wrong person. The person willing to take a call with an early-stage founder is not a random sample of your target market. They are disproportionately curious, startup-sympathetic early adopters. Their feedback is not representative of the broader segment you need to sell.
You ask for opinions instead of evidence. Opinions are cheap and often wrong. Evidence is specific, behavioral, and grounded in past experience. "What do you think about X?" generates opinion. "Tell me about the last time you had to deal with X" generates evidence.
The Five-Question Interview
These five questions, asked in this order, generate more actionable signal than fifty generic discovery calls. None of them describe your product. All of them are about the customer's past behavior.
Question 1: "Tell me about the last time you had to deal with [the problem you are solving]. Walk me through what happened."
This is the opening. You are looking for a specific story, not a general opinion. Probe for: who was involved, what tools they used, what the outcome was, how long it took. If they cannot tell you about a specific recent instance, the problem may not be as real or frequent as you assumed.
Question 2: "What did you do to try to solve it?"
This reveals the existing solutions in the market. You will learn more about your competitive landscape from this question than from any market analysis. If they say "nothing," the problem is either not painful enough to warrant action or they have tried and failed so many times that they have given up. Both are important findings.
Question 3: "How much did that cost you? Time, money, or both?"
This is the willingness-to-pay signal in disguise. If they can quantify what the problem costs them, they can value a solution. Vague answers here are worth noting: they may mean the problem is real but episodic, which has implications for how you price.
Question 4: "What would a perfect solution have done that the current approach does not?"
This is the only forward-looking question and it is deliberately placed after three questions that have grounded the customer in their real experience. The answers here are far more useful than answers to the same question asked cold, because now the customer is describing gaps in solutions they have actually used rather than features they imagine would be nice.
Question 5: "Who else in your organization deals with this? Would they ever buy a solution together?"
This is the champion-mapping question. At a B2B startup, the person you are talking to is almost never the only buyer or the only user. Understanding the buying group and the internal dynamics around this problem is critical before you design your sales motion. A tool your contact loves but their manager would never approve is a different product than one that solves the manager's problem first.
Running the Interview
The mechanics matter as much as the questions.
Do not record without explicit permission, but do take notes during the call. The exact words customers use to describe their problems are your marketing copy. "Our reports are always wrong when we hand them to the client" is a better headline than anything you will write in a copywriting session.
Stay quiet after each question. Five seconds of silence will often produce the most important part of the answer. Founders who are uncomfortable with silence fill it, and what they fill it with is usually their own hypothesis. Resist.
Time-box to 25 minutes. Longer calls lose energy. If you cannot get usable signal in 25 minutes, the interview structure is the problem, not the duration.
Run a minimum of fifteen interviews before drawing conclusions. Five is too few to see patterns. Fifteen is the minimum where themes start to emerge across different people and contexts.
The Decision Tree for Interview Findings
After fifteen interviews, categorize what you learned using this framework:
Did 10+ of 15 describe the same specific problem without prompting?
Yes: You have a real problem worth solving.
No: The problem may be a subset of a different, larger problem.
Did 8+ actively try (and fail) to solve it with existing tools?
Yes: There is a pull market. People want a solution.
No: You may need to create demand, not just fill it.
Could 6+ quantify the cost in time or money?
Yes: Pricing discovery is possible in your next round of interviews.
No: The problem is real but may not be acute enough to command budget.
Did 5+ mention the same missing capability in existing solutions?
Yes: That capability gap is your product wedge.
No: Your differentiation needs more investigation before you build.
If you answer yes to all four, you have genuine product-market signal. If you answer yes to fewer than three, you either have the wrong customer segment, the wrong problem framing, or both. Go back to interviews before writing a line of code.
What I Got Wrong
At my first startup, I ran what I called "customer development" for three months and came away convinced we had strong product-market fit. I had talked to forty people. I had hundreds of notes. I had quotes I could use in board meetings.
What I actually had was forty polite conversations with people who thought our idea sounded interesting. I had asked every one of them some version of "would you use this?" I had not asked a single one of them to tell me about the last time they actually had the problem. I was measuring enthusiasm for a pitch, not demand for a solution.
We launched. We got 200 signups in the first week. Twelve converted to paid. The rest disappeared. I went back through every customer development note and could not find a single piece of evidence that would have predicted that conversion rate. Because there was none. There was only opinion.
The Anti-Patterns to Watch
The customer who loves every idea. They are helpful. They want you to succeed. They are not a signal of anything except their own optimism.
The expert who wants to consult. They will give you a 45-minute framework for how the industry works, none of which answers the five questions above. Redirect them: "That context is really useful. Can I ask you about a specific situation you were in recently?"
The first-time storyteller. They describe the problem as it happened once, in an unusual situation. You need the recurring, typical instance. Ask: "Is that typical of how this usually happens? How often does this come up?"
Your warm network. Friends-of-friends who take calls because they know you are predisposed to like your idea. Reserve them for a later round of concept validation. Use cold outreach for the first fifteen interviews. Cold interviews are more honest because the interviewee has no social reason to be generous.
What Good Discovery Output Looks Like
After fifteen interviews done correctly, you should be able to write a two-page document that answers these questions without hedging:
- The specific situation in which this problem occurs (trigger, frequency, who is involved)
- The tools or processes they currently use to address it (competitive landscape, not guessed)
- The exact language they used to describe the problem (verbatim, three to five quotes)
- The quantified cost of the problem for the respondents who could quantify it
- The one or two capability gaps mentioned most often across different respondents
If you cannot write that document after fifteen interviews, you need fifteen more or you need to revisit your interview structure. Discovery is an information-gathering exercise, not a validation exercise. The fastest way to accelerate your startup is to structure it like one.